Technical Notes Advances in Structural Engineering 1–15 Ó The Author(s) 2019 Article reuse guidelines: sagepub.com/journals-permissions DOI: 10.1177/1369433219849829 journals.sagepub.com/home/ase Faster multi-defect detection system in shield tunnel using combination of FCN and faster RCNN Xinwen Gao 1,2 , Ming Jian 1 , Min Hu 2,3 , Mohan Tanniru 4 and Shuaiqing Li 1 Abstract With the large-scale construction of urban subways, the detection of tunnel defects becomes particularly important. Due to the com- plexity of tunnel environment, it is difficult for traditional tunnel defect detection algorithms to detect such defects quickly and accu- rately. This article presents a deep learning FCN-RCNN model that can detect multiple tunnel defects quickly and accurately. The algorithm uses a Faster RCNN algorithm, Adaptive Border ROI boundary layer and a three-layer structure of the FCN algorithm. The Adaptive Border ROI boundary layer is used to reduce data set redundancy and difficulties in identifying interference during data set creation. The algorithm is compared with single FCN algorithm with no Adaptive Border ROI for different defect types. The results show that our defect detection algorithm not only addresses interference due to segment patching, pipeline smears and obstruction but also the false detection rate decreases from 0.371, 0.285, 0.307 to 0.0502, respectively. Finally, corrected by cylindrical projection model, the false detection rate is further reduced from 0.0502 to 0.0190 and the identification accuracy of water leakage defects is improved. Keywords cylindrical projection, deep learning, faster RCNN, field of view conversion, FCN, multi-defect of tunnel detection, ROI Introduction Urban subway tunnels suffer from cracks, water leak- age, shelling and faulty platforms over time due to complex geological conditions, climate change, and evolving design and maintenance of construction tech- nology. If these problems are not detected and repaired in time, they can lead to tunnel lining (i.e. a wearing of tunnels) and other potential hazards, as well as contribute to safety challenges in the urban rail transit systems. Traditional detection of defects relies on inspecting the tunnels manually and recording the images for later observation. This type of observation and analysis of defects is subjective and can lead to erroneous judgements, omissions and other types of mistakes. Also, such an analysis is time-consuming, labour intensive and inefficient. In recent years, image processing detection methods using computer vision technology have gained promi- nence (Huang et al., 2017; Zhang et al., 2014). These detection methods analyse large amount of image data of tunnel structures, but becomes extremely compli- cated due to interference at multiple tunnel shields, joints and pipelines, paint numbers and so on, each contributing to errors in detecting defects. These errors increase over time and, if not detected, can quickly reduce the strength of concrete used in tunnel lining and contribute to damage of the tunnel structure. Hence, there is a need for a fast, reliable and efficient method to detect defects in large-scale urban subways. Protopapadakis and Doulamis (2015) extracted 17 low-level features such as grayscale, edge, frequency, entropy, texture and HOG (Makantasis et al., 2017; Protopapadakis and Doulamis, 2015) from the original tunnel image and used these features as input into the CNN (convolutional neural network) to obtain advanced features, which are then used as input to 1 Institute of Mechanical and Electrical Engineering and Automation, Shanghai University, Shanghai, China 2 SHU-SUCG Research Center of Building Industrialization, Shanghai University, Shanghai, China 3 SILC Business School, Shanghai University, Shanghai, China 4 Oakland University, Rochester, MI, USA Corresponding author: Xinwen Gao, Institute of Mechanical and Electrical Engineering and Automation, Shanghai University, Shanghai 200444, China. Email: gxw3405@shu.edu.cn